Semantic-Vision-Public

Semantic Vision - AI Accessibility Layer for DeFi

Zero-Integration. Zero Protocol Changes. Full Accessibility.

Status: Private Alpha License: MIT Python 3.12 Powered by Groq Chrome Extension MV3 EU Accessibility Act 2025

The Manifesto

I lost my sight at fourteen.

At twenty-two, I opened a DeFi lending protocol for the first time. I heard this:

“Niezidentyfikowany. Niezidentyfikowany. Niezidentyfikowany.”

Fourteen buttons. Fourteen times: Unidentified. No context. No way to know which one supplies ETH, which one borrows USDC, which one could empty my wallet if pressed by mistake.

Decentralized finance promises permissionless access to capital for everyone. That promise has an asterisk — and the asterisk says: unless you are blind.

I am blind. I refused to accept that asterisk. So I built the solution myself.

Semantic Vision is an AI layer that reads DeFi interfaces and generates meaningful ARIA labels in real time — on any protocol, without touching a single line of their code. It works today, on protocols that will never run an accessibility audit, for users who cannot wait.

This is not a charity project. This is infrastructure.

— Vladyslav, Founder

The Problem

The DeFi ecosystem manages over $100 billion in assets. It is built on the principle of open, permissionless access.

Yet for 2.2 billion people worldwide with visual impairments — including 43 million who are blind — decentralized finance is a wall of silence.

What a sighted user sees What a blind user hears
“Supply ETH” button “Niezidentyfikowany” (Unidentified)
“Borrow USDC — 3.2% APY” button “Niezidentyfikowany”
“Connect Wallet” button “Niezidentyfikowany”
“Close modal” button “Niezidentyfikowany”
14 interactive controls on Aave 14 × “Niezidentyfikowany”

Existing solutions all require protocol developers to act:

The result: blind users are systematically excluded from an entire financial paradigm. Not by malice. By indifference and inertia.

The Solution

Semantic Vision is a Zero-Integration AI Accessibility Overlay.

It sits between the browser and the screen reader. It reads the DOM. It understands DeFi. It injects accurate ARIA labels in real time — without the protocol doing anything.

  Without Semantic Vision With Semantic Vision
Setup required from protocol — None
Aave v3 buttons labeled 0 / 14 14 / 14
“Unidentified” announcements 14 0
Modal focus management Broken Correct
Dynamic elements (modals, dropdowns) Unlabeled Labeled in real time
Works offline (cached) — ✅ 7-day cache

Why Zero-Integration changes everything

Every other accessibility solution requires someone on the protocol team to write code, run audits, and maintain compliance. That requires budget, priority, and time — none of which blind users control.

Semantic Vision requires nothing from protocols. It deploys client-side, runs locally on the user’s machine, and works on any web interface. The same architecture that labels Aave’s buttons today can label any site tomorrow — DeFi protocols, Web2 applications, e-commerce platforms. Adjusting the configuration and key bindings is sufficient to target a new domain entirely.

This is not a DeFi-only product. It is a universal accessibility layer that starts with DeFi.

Visual Proof — Aave v3

Screenshot 1 — Before Semantic Vision

Image description: NVDA screen reader speech log on app.aave.com showing entries reading “Niezaetykietowany” (Polish for “Unlabeled”) for interactive buttons on the Aave lending interface. The user cannot determine which button is Staking menu, Send token, or Receive token. No accessible name exists on any interactive control.

Before: NVDA reads "Niezaetykietowany" multiple times on Aave v3. Not all buttons have an accessible name.

Screenshot 2 — After Semantic Vision

Image description: NVDA screen reader speech log on the same app.aave.com page after Semantic Vision activation. Interactive elements now have meaningful labels: “Staking menu button”, “Send token button”, “Receive token button”. The aria-live region has announced “elements labeled successfully.”

After: Aave v3 buttons labeled - Staking menu, Send token, Receive token. Zero unidentified elements.

Live Demo

Private Beta Demo available upon request for strategic partners and investors.

To schedule a live demonstration via video call: → Contact via Twitter/X DM

Demo overview: Screen recording showing NVDA navigating app.aave.com. First segment: without the extension — buttons announce “Niezidentyfikowany”, navigation is impossible. After Semantic Vision activates: buttons are correctly labeled, the modal opens with proper focus management, and the aria-live region confirms successful labeling.

Architecture

Semantic Vision runs entirely on the user’s machine. No wallet addresses, transaction data, or private keys ever leave the device. Only DOM structure and visual bounding boxes are sent to the AI inference endpoint.

┌─────────────────────────────────────────────────────────┐
│                 Chrome Extension (MV3)                  │
│                                                         │
│   content.js              background.js                 │
│   ┌───────────────┐       ┌─────────────────┐          │
│   │ DOM Collector │──────▶│ Native Messaging │          │
│   │ MutationObsvr │       │ Relay           │          │
│   │ ARIA Injector │◀──────│                 │          │
│   │ Focus Manager │       └────────┬────────┘          │
│   │ aria-live     │                │                    │
│   └───────────────┘                │                    │
└───────────────────────────────────│────────────────────┘
                                    │ Native Messaging
                                    ▼
                       ┌────────────────────────┐
                       │    native_host.py       │
                       │    Windows Registry     │
                       │    Bypasses Chrome PNA  │
                       └────────────┬───────────┘
                                    │ HTTPS (mkcert)
                                    ▼
                       ┌────────────────────────┐
                       │   server.py — FastAPI  │
                       │   localhost:8000        │
                       └──────────┬─────────────┘
                                  │
                    ┌─────────────┴─────────────┐
                    ▼                           ▼
        ┌──────────────────┐       ┌─────────────────────┐
        │ SpatialClusterer │       │    GroqClient        │
        │ grid=300px       │       │    LLaMA 3.3 70B     │
        │ max_dist=250px   │       │    + fallback chain  │
        └──────────────────┘       └─────────────────────┘

Diagram description: Chrome Extension collects DOM elements and routes them via Native Messaging to a local FastAPI server. The spatial clustering algorithm groups nearby elements by proximity, providing richer context to the LLaMA 3.3 70B model via Groq API. Generated labels are injected back into the live DOM as aria-label attributes. The full chain runs locally on the user’s machine.

Tech Stack

Layer Technology
Extension Chrome Manifest V3, Native Messaging API
Backend Python 3.12, FastAPI, Pydantic v2
AI Model LLaMA 3.3 70B via Groq API
AI Fallback OpenRouter, Gemini
DOM Parsing Playwright
Clustering Custom grid-based spatial proximity algorithm
SSL mkcert (local certificate authority)
Cache chrome.storage.local, 7-day TTL
Platform Windows 10+, NVDA (JAWS: Stage 2 roadmap)

Tested Protocols

Protocol Elements Labeled “Unidentified” After Status
Aave v3 14 / 14 0 ✅ Tested
PancakeSwap Partial — 🔄 Early testing — Stage 2
Uniswap v3 — — 📋 Stage 2
Curve Finance — — 📋 Stage 2
Compound — — 📋 Stage 2
Web2 / E-commerce — — 📋 Stage 2 (configuration change only)

Access & Installation

Semantic Vision is currently in private alpha.

The core architecture is proven and working. Public deployment packaging — one-click installer, Docker container, automated SSL setup — is on the Stage 2 roadmap alongside the first funding round.

The current build is available for review and live demonstration to strategic partners and investors. Easy-install scripts and full public documentation are planned for Stage 2.

→ Request access or demo via Twitter/X DM

Roadmap

Stage 1 — Foundation ✅ Now

Stage 2 — First Funding Round 🎯

Stage 3 — European Market

Stage 4 — Global Accessibility Infrastructure

Grant Applications & EU Compliance

The EU Web Accessibility Act 2025 (European Accessibility Act, Directive 2019/882) mandates accessible digital financial services across all EU member states from June 2025. DeFi protocols operating in the EU face increasing regulatory pressure to meet these standards.

Semantic Vision is the only solution that achieves compliance without requiring protocol-side development work. This creates a direct alignment with the objectives of Polish and European innovation funding programs targeting digital inclusion and regulatory readiness.

Program Focus Potential Amount Status
Aave Grants DAO Direct protocol accessibility TBD 🎯 Applying
Gitcoin Grants DeFi public goods Community-matched 🎯 Applying
EIC Accelerator European deep tech Up to €2.5M 🎯 Preparing
NCBR / PARP Polish national R&D TBD 🎯 Preparing
KPT Kraków Innovation Hub Regional incubation TBD 🤝 In talks

For Investors & Strategic Partners

“The DeFi accessibility gap is not a niche problem. It is infrastructure debt sitting beneath $100B+ in assets — and right now, it has no solution that works without protocol cooperation.”

Semantic Vision is the first AI-powered Zero-Integration accessibility layer built specifically for decentralized finance, created by a blind founder who is himself the primary user.

What makes this defensible:

We are in private alpha and actively seeking our first strategic funding round to assemble the team and expand protocol coverage.

→ Start a conversation via Twitter/X DM

Security & Privacy

License

MIT License — see LICENSE for details.

Acknowledgements

Semantic Vision — Because financial freedom shouldn't require perfect vision.

GitHub: @Wislaw · Twitter/X: DM open